多先验驱动的单图像联合去马赛克及超分辨率技术研究

Due to limitation of sensor resource, image demosaicking and super resolution techniques have broad application prospects. However, the two techniques are usually separately performed, which may easily lead to the spread of demosaicking errors in the reconstructed high resolution (HR) images. In order to solve this problem, we will study the single image joint demosaicking and super resolution technique, which aims to directly recover the high quality HR images from the measurements sampled by the color filter array. We expect that our proposed technique could well preserve the edges and details in natural images, while generate results without noticeable artifacts such as false color, zipper effect, aliasing, ringing, etc. The main research contents are listed as follows: 1) the collaborative-representation-based modeling of external full color HR prior information; 2) the statistical modeling of internal prior information which incorporates local similarity, non-local similarity and inter-channel correlation of color images; 3) establishing the reconstruction model which combines the external and the internal priors and designing the effective algorithms for the optimization problem. The research of this project is especially useful for different applications where the sensor resource is highly limited. Moreover, it can promote the developments and the applications of the super resolution and the demosaicking techniques, and also offer theoretical and technical support for other related research fields.

由于传感器资源受限，图像去马赛克技术及超分辨率技术具有广阔的应用需求。但是，现有的技术多是将二者分离处理，因此容易引起去马赛克误差在重建高分辨率图像中扩散。为了解决该问题，本项目研究单图像联合去马赛克及超分辨率技术，目标是直接从色彩滤波器阵列欠采样图像中恢复出高质量的高分辨率图像，在最大限度地保持重建图像边缘锐利、细节完整的同时防止伪色彩、拉链效应、混叠、振铃等假象的出现。主要研究内容部包括：1）研究基于协作表达的外部全彩高分辨率先验信息建模方法；2）研究融合彩色图像局部、非局部相似性及通道间相关性的内部先验统计建模方法；3）建立融合外部及内部联合先验的重建模型并研究高效的求解算法。本项目的研究，特别适用于传感器资源受限的各类应用场景，可以推动去马赛克、超分辨率技术的进一步发展与应用，并可为相关研究领域提供理论参考与技术支撑。

由于传感器资源受限，图像去马赛克技术及超分辨率技术具有广阔的应用需求。但是，现有的技术多是将二者分离处理，因此容易引起去马赛克误差在重建高分辨率图像中扩散。为了解决该问题，本项目研究单图像联合去马赛克及超分辨率技术，目标是直接从色彩滤波器阵列欠采样图像中恢复出高质量的高分辨率图像，在最大限度地保持重建图像边缘锐利、细节完整的同时防止伪色彩、拉链效应、混叠、振铃等假象的出现。主要研究内容部包括：1）研究基于协作表达的外部全彩高分辨率先验信息建模方法；2）研究融合彩色图像局部、非局部相似性及通道间相关性的内部先验统计建模方法；3）建立融合外部及内部联合先验的重建模型并研究高效的求解算法。本项目的研究，特别适用于传感器资源受限的各类应用场景，可以推动去马赛克、超分辨率技术的进一步发展与应用。经过四年的研究，本项目组发表学术论文14篇，其中SCI检索5篇；获授权中国发明专利4项，另有2项中国发明专利在实质审查状态；申请并获得软件著作权登记6项；培养硕士研究生9人。